The present invention relates generally to image processing, and more particularly, but not by way of limitation, relating a system, a method, and a recording medium for compressing aerial videos.
Bandwidth for transmitting airborne video imagery to ground is limited and is often significant bottleneck for example, viewing the video as, drones are capturing the video. The available bandwidth is significantly lower than the data capture rate. Therefore, conventional compression methods are not suitable.
Conventional methods have used subsampling with temporal and spatial data, MPEG4 using spatial, temporal, and channel redundancies, on demand and operator initiated, analytics based on where event summaries are sent, and stitching to send a panorama. However, such conventional compression methods do not provide enough details, enough compression, only provide information in real time, provide all information in the videos such that significant bottleneck is created, and the stitching has stitching-based artifacts, for example, due to parallax.
Other conventional methods have detected a region of interest, but such methods are only good for static cameras. Some conventional methods use background data to improve the data compression ratio, but such methods do not work well for a camera mounted on a moving platform (e.g., a drone).
Even further, another conventional method has proposed to use global motion (homography) to initialize the block matching algorithm, but this method does not further improve a data compression ratio. Also, a conventional method has proposed to use homography to model global motion and to estimate the intermediate frames within a same frame group. The aforementioned method is limited by the frame group sizes since the method cannot have a large frame group.
All these conventional methods are limited in their applications. For example, the methods, if utilized in a moving camera, do not compress well because of the frame to frame motion. The conventional methods do not explicitly recover this motion and hence may not deliver high compression of aerial videos.